mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-29 06:33:28 +00:00
35753be11f
FlexAI's Token Service (https://api.flex.ai/v1) is an OpenAI-compatible inference endpoint serving open-weight models. This registers it as the `flexai` provider via the JSON-based openai_like configuration. - litellm/llms/openai_like/providers.json: `flexai` entry, base_url https://api.flex.ai/v1, FLEXAI_API_KEY, FLEXAI_API_BASE override, /v1/chat/completions + /v1/responses - litellm/constants.py: `flexai` in openai_compatible_providers and https://api.flex.ai/v1 in openai_compatible_endpoints, so JSONProviderRegistry.get_by_base_url resolves an api_base override - litellm/types/utils.py: FLEXAI in the LlmProviders enum - litellm/proxy/public_endpoints/provider_create_fields.json: Add Model dropdown entry (api_key required, api_base optional) - model_prices_and_context_window.json (+ the packaged backup copy): 28 priced entries covering the full served roster — 23 chat, 1 embedding, 1 image, 1 speech, 2 transcription — with per-model cache-read pricing so cost tracking stays accurate under prompt caching - provider_endpoints_support.json: endpoint-support block - tests/test_litellm/llms/openai_like/test_flexai_provider.py: 18 tests Token pricing comes from FlexAI's GET /v1/models; per-image, per-character and per-minute media rates come from GET /v1/pricing, which is the only endpoint carrying them. The two agree on every chat row. Context windows come from FlexAI's serving catalog rather than the /v1/models metadata, which under-reports on some models (a 70,043-token prompt is accepted by a model whose metadata advertises 65,536). Non-chat entries use the cost fields the generic calculators actually read (`input_cost_per_image`, `input_cost_per_character`, `input_cost_per_second`); the output_* variants are ignored by those calculators and raise instead. Regression tests exercise the image and speech paths. No provider-specific branch is needed in get_llm_provider_logic.py: the generic JSON-provider lookup resolves flexai from providers.json by base URL, and already prefers an explicitly supplied api_key over the server environment. The tests cover that path. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| _support | ||
| audio_tests | ||
| base_sdk_tests | ||
| benchmarks | ||
| code_coverage_tests | ||
| documentation_tests | ||
| e2e | ||
| e2e_harness | ||
| fixtures | ||
| harness_e2e | ||
| integration | ||
| rust-python-harness | ||
| test_litellm_rust | ||
| typing | ||
| unit | ||
| windows_tests | ||
| __init__.py | ||
| _fake_openai_endpoint_server.py | ||
| _flush_vcr_cache.py | ||
| _live_test_helpers.py | ||
| _master_key.py | ||
| _openai_record_replay_proxy.py | ||
| _process_helpers.py | ||
| _vcr_conftest_common.py | ||
| _vcr_redis_persister.py | ||
| _wait_helpers.py | ||
| _ws_vcr.py | ||
| AGENTS.md | ||
| capturing_transport.py | ||
| eval_swe_bench.py | ||
| fake_openai_endpoint.py | ||
| gettysburg.wav | ||
| large_text.py | ||
| openai_batch_completions.jsonl | ||
| pyrightconfig.json | ||
| README.MD | ||
| test_logging.conf | ||
| test_rust_python_harness.py | ||
| VCR_TESTING.md | ||
| white_100x100.png | ||
In total litellm runs 1000+ tests
[02/20/2025] Update:
To make it easier to contribute and map what behavior is tested,
we've started mapping the litellm directory in tests/unit
This folder can only run mock tests.